Date of Award
2025-05-01
Degree Name
Master of Science
Department
Computer Science
Advisor(s)
Martine Ceberio
Abstract
Recent advancements in machine learning have led to the design of many neural network architectures aimed at solving real-world problems. Each network works to make predictions by finding patterns in the provided data. One common application is time series forecasting, where a model predicts future events based on historical time series data. Time series forecasting is used in a variety of fields, one example being in predicting water releases of Hybrid Floating Photovoltaic-Hydropower (HFPVH) systems. As global population growth drives an increase in energy demand, the need for resilient and sustainable energy generation has become urgent. HFPVH systems have emerged as a promising renewable energy source to help with the increase in energy demand. This master's thesis explores modifying a time series forecasting technique known as Long Short-Term Memory (LSTM). As machine learning has evolved, networks have continuously grown in size, resulting in better predictions. However, this commonly comes with trade-offs in longer training time, energy consumption, and hardware constraints. These trade-offs highlight the importance of a lightweight LSTM architecture that maintains or enhances the predictive accuracy of the standard LSTM model while maintaining a smaller network size. Furthermore, the proposed LSTM architecture integrates a dynamic outlier filter that aims to maintain the same data flow through the LSTM but enhance the convergence of the model's predictions, reducing the need for deeper architectures. The improved architecture is tested across multiple forecasting scenarios, including energy consumption rates, pollution levels, Apple stock prices, and HFPVH systems, showcasing improved accuracy and efficiency compared to the standard LSTM model. In addition to forecasting, this thesis develops a modular constraint-based system that ensures that policies for HFPVH are maintained. Its modularity offers the flexibility of interchanging different functions or models in the simulation, enabling reliable performance in a wide range of climates.
Language
en
Provenance
Received from ProQuest
Copyright Date
2025-05
File Size
82 p.
File Format
application/pdf
Rights Holder
Jose Reynaldo Vega
Recommended Citation
Vega, Jose Reynaldo, "Machine Learning And Time Series Forecasting For Hydropower Predictions" (2025). Open Access Theses & Dissertations. 4493.
https://scholarworks.utep.edu/open_etd/4493